The tracking of surgical instruments o fers interesting possibilities for the development of high-level commands for robotic camera holders in laparoscopic surgery. We have developed a new method to detect instruments in laparoscopic images which uses information on the 3D position of the insertion point of an instrument into the abdominal cavity. This information strongly constrains the search for the instrument in each endoscopic image. Hence, the instrument can be detected in near real-time using shape considerations. Early results on laparoscopic images show that the method is rapid and robust in the presence of partial occlusion and smoke. Our first experiment on a cadaver validates our approach and shows encouraging results